3.1 Polychlorinated biphenyls, dibenzofurans and mortalities
The follow-up study of mortality status has been approved by the Institutional Review Board of the National Taiwan University Medical Center. We began with the Yucheng Registry (Hsu et al., 1985) and the list of neighborhood referents that was described previously (Yu et al., 1997; Guo et al., 1999). When the PCB poisoning was discovered in 1979, a registry of subjects was set up by the Taiwan Provincial
Department of Health. The criteria used to identify subjects included consumption of brands of rice oil produced in the factory known as the source of the contamination, and the development of skin, nail, eye, and other symptoms from January to October 1979.
About 10% of persons in the registry gave a history of PCB exposure and had elevated serum PCB concentrations but were asymptomatic (Hsu et al., 1985).
A total of 2,061 subjects were included in the Yucheng registry by 1983, including 70 children born to exposed mothers. Among the 1,991 in the list who were directly exposed, 154 did not have an address, and thus could not be traced further. These individuals were excluded from the first study of mortality in 1992 (Yu et al., 1997).
Each person in Taiwan is assigned a national identification number when registering their first permanent address. We had the national identification number for all
participants. A typical national identification number has 1 English letter and 9 digits.
The last digit of the national identification number is a check digit, which is generated from the letter and the other digits based on a formula. The check digit helps identify legitimate national identification numbers. Among the national identification numbers listed for the Yucheng subjects, 34 had check digits that were inconsistent with the rest
13
of the number and were excluded. We thus had 1,803 exposed Yucheng subjects for this study.
The neighborhood referents were recorded as residents, in 1979, of the same community (usually on the same street), had the same gender, had birthdays within 3 years of the Yucheng subjects’ birthday, and were not themselves in the registry. We attempted to identify three referents for each Yucheng subject, and found 5,519 eligible persons. Among them, 8 had no date of birth recorded, and 98 were born after June 30, 1978, during or after the incident, and so had been selected in error and were excluded.
An additional 243 had national identification numbers with an inconsistent check digit, and they were excluded, leaving 5,170 neighborhood referents. The final ratio of Yucheng subjects to neighborhood referents was 1: 2.9. We previously reported that these referents were of similar education, occupation, and socioeconomic status as the Yucheng subjects (Guo et al., 1999).
We compared the national identification numbers of these two groups with the national mortality registry to determine vital status, and, if deceased, date of death and cause of death. In Taiwan, it is mandatory to report deaths within 1 month to the local registration office responsible for the area in which the deceased resided. These reports are believed to be complete. The cause of death was coded according to the ninth revision of the International Classification of Diseases (ICD-9). Unfortunately, the national mortality registry did not contain national identification numbers until 1985. To address this issue, we used another mortality registry to obtain identification numbers.
The Taiwan Ministry of the Interior had established a separate death registry based on death certificates. This registry had complete identification numbers, but lacked the ICD-9 codes. We submitted the identification numbers to the Ministry of Interior’s
14
registry to determine fact and date of death, and then, using them plus the date of birth and area code of the decedents, matched to the national mortality registry for ICD-9 cause of death. We only accepted exact matches. This yielded a valid match for 89% of all decedents from 1980-84.
Among the 1,803 Yucheng subjects and 5,170 neighborhood referents, 295 and 757 had died between January 1, 1980 and December 31, 2008, respectively. The overall and cause-specific mortality of the Yucheng subjects were compared to neighborhood
referents. Each study subject contributed observed person-time from January 1, 1980 to the date of the end of follow-up (December 31, 2008) or through their date of death.
We used an age-stratified method to calculate standardized mortality ratio (SMR).
The person-years at risk for all subjects were combined into gender, 5-year age, and 1-year calendar time-specific groups. The accumulated person-1-years were then multiplied by the gender, age, calendar time, and cause-specific neighborhood referents’ mortality rates to yield the expected numbers of cause-specific deaths. The observed number of cause-specific deaths was then divided by the expected number of cause-specific deaths to yield the SMR. The 95% confidence intervals around the SMRs were estimated based on the Fisher mid-P exact confidence intervals. Mortality data also were analyzed separately by gender. For those national identification numbers that appeared in the mortality database, but lacked an ICD-9 code, the person-time of observation was still included when calculating SMRs.
15
3.2 Mortalities among Yucheng and Yusho victims
Data extraction
The articles and the extracted data were reviewed independently by two
investigators (Ming-Chieh Li and Yue-Liang Leon Guo). The following information was
extracted from the two studies: first author, year of publication, country, cohort size,
follow-up period, person-year of follow-up, and SMRs and 95% CI for cancer or
non-cancer diseases. The cause of death was coded according to the ninth revision of the
International Classification of Diseases (ICD-9). Information on chemical contaminants
in the oils were extracted from Masuda et al., 1986.1 Exposure levels were extracted
from Guo et al., 19979, Lambert et al., 2006,17 Tanabe et al.,18 1989, and Masuda et al.,
1998.19
Statistical analysis
Log-transformed 95% CIs were calculated for estimating the standard errors (SEs)
for the ln(SMR) by the formula: SE=[ln(upper limit)-ln(lower limit)]÷(2×1.96).20 For
diseases in which there was no death (0) observed, we elected not to calculate the
combined SMRs.
16
Overall pooled SMRs and 95% CIs were obtained using random-effects methods.21 Heterogeneity of effects across studies was assessed by the Cochran’s Q statistic and
was deemed significant when p-Value < 0.10. To evaluate the percentage of variation
attributable to heterogeneity, we calculated the I2 statistics.22 Substantial heterogeneity
was defined at I2 > 50% and/or p-Value <0.10 by the Q test. For those diseases with
significant heterogeneity, summary mortality was not calculated, and only results from
individual cohorts were shown. Statistical analyses were conducted using STATA 12.0
(Stata Corp, College Station, TX), notably the metan commands. Statistical significance
of pooled SMRs was defined as a p-Value of lower than 0.05.
17
3.3 Gestational exposure to polychlorinated biphenyls and hearing loss
This study was approved by the Institutional Review Board of The National Taiwan
University Medical Center. The study candidates were from two groups, the exposed
Yucheng children who were born between June 1978 and December 1998 to mothers
exposed to PCBs and PCDFs (Chen et al. 1992; Guo et al. 1994). The other group was
from the previously identified referent group of children. For each Yucheng child, one
unexposed child was selected as a control by matching for neighborhood (lived /born
within the same township), age (within 15 days of age for those under 1 year, and within
1 month of age for those older), gender, mother's age (within 3 years of age), parents'
combined educational level (within 3 years), and occupation (within 1 class of 5 classes
from unskilled laborer to professional). A total of 240 exposed and 240 unexposed were
entered into follow-up. In 2007, a health survey was conducted in three townships,
where a total of 184 Yucheng children and 184 referent children were found and invited
to participate in a health examination. An informed consent was obtained before
examination and tests.
Methods of measurements
Demographic data were collected by using a structured questionnaire. Otoscopic ear
18
examination was carried out by an otolaryngologist (H. P. Wu) before hearing tests.
Participants with ear disease or other pathologies related to hearing loss were excluded.
Blood samples from all children were collected for measurements of serum
concentrations of cholesterol and triglycerides. All interviewers, physicians, and testers
were blinded as to the exposure status of the participants.
Pure tone audiometry
Pure tone audiograms were obtained for each ear in all subjects, employing a
standard threshold search procedure using a clinical audiometer (Unity PC Audiometer
SD 100, Copenhagen, Denmark). Pure tone thresholds were obtained from 250 to 8000
Hz, via headphones. All pure tone threshold tests were conducted in a sound-proofed
booth.
Distortion Product Otoacoustic Emissions testing
Distortion product otoacoustic emissions (DPOAEs) were elicited by two continuous, primary tones at frequencies ƒ1 and ƒ2 (ƒ2 : ƒ1 was fixed at 1.22),
19
generated by two separate transducers (ER2; Etymotic Research, Elk Grove Village, IL,
USA) connected to a digital signal processing board (DSP) (SmartOAE; Intelligent
Hearing Systems, Miami, FL, USA). Two short plastic tubes connected the transducer
outputs to the OAE probe (ER10B+; Etymotic Research), which also contained a
miniature low-noise microphone for emission detection. The tapered end of the probe
was extended with a short, soft silicon tip and the probe was inserted deeply and tightly
into the external ear canal. The microphone detected the overall acoustic signal in the
external ear canal during tone stimulation. Its response was preamplified (+40 dB) then
analyzed by the SmartOAE DSP board. The stimulation and detection process were
automatically controlled by using a PC computer driven by SmartOAE software
(Intelligent Hearing Systems, version 3.72). This software, based on Fourier
transformation calculation, generated the stimulations through two independent
channels, checked the actual levels of stimulating tones, and computed the complex response at frequency 2ƒ1- ƒ2. The testing frequency range of ƒ2 was 1.5-6 kHz, and
the two primaries, L1 and L2, were set at 65 and 55 dB SPL. Amplitudes of DPOAEs were measured at selected frequencies. Data are described with respect to ƒ2 frequency
since the generator site of the 2ƒ1-ƒ2 distortion product has been most closely
correlated with the ƒ2 frequency place in the cochlea.
20
Maternal exposure data
Due to a broad coverage of media about this event, all the women who were
followed ceased using rice oil. Since the government halted the manufacturing and sale
of the contaminated oil, further exposure was unlikely. However, the toxic compounds
were known to have long body half-lives, and Yucheng mothers with a high body
burden continued to transfer the toxic substances to their fetuses even several years
later.
Maternal blood concentrations of PCBs and PCDFs were available from 53 children
of 47 mothers. The blood samples were collected between 1994 and 2003, stored at -80
°C, and sent on dry ice to the U.S. Centers for Disease Control and Prevention for the
measurements of PCB, PCDF, and PCDD congeners (Lambert et al., 2006). In brief,
high resolution gas chromatography (GC) / mass spectrometry (MS) (Patterson et al.,
1987) was used for the measurement of 2,3,4,7,8-PeCDF, 1,2,3,4,7,8-HxCDF.
Ortho-substituted PCBs were analyzed by a Hewlett-Packard 5890 gas chromatograph (GC)
(Hewlett-Packard, Houston, TX) using an electron-capture detector, including
2,3’,4,4’,5-PeCB (IUPAC118), 2,2’,4,4’,5,5’-HxCB (IUPAC153), 2,2’,3,4,4’,5’-HxCB
(IUPAC138), 2,3,3’,4,4’,5-HxCB (IUPAC156), 2,2’,3,3’,4,4’,5-HpCB (IUPAC170),
2,2’,3,4,4’,5,5’-HpCB (IUPAC180) Values were reported on a lipid weight basis in parts
21
per trillion (ppt) by dividing the congeners on a whole-weight basis by total serum lipid
content, estimated from measurements of triglycerides, and total cholesterol (Phillips et
al. 1989).
Yucheng children were born between June 1978 and December 1998, i.e., earlier
than the time we got maternal exposure data. Thus, gestational exposure was estimated
by back-extrapolation from the mothers’ serum concentrations. The half-life used for
back extrapolation of each congener was based on a review article, which was specific
for Yucheng and Yusho cohorts (Ogura, 2004), namely 1.7 years for PCB-118, 3.9 years
for PCB-153, 4.8 years for PCB-138, 4.9 years for PCB-156, 5.4 years for PCB-180, 5.5
years for PCB-170, 3.1 years for 2,3,4,7,8-pnCDF, and 3.3 years for 1,2,3,4,7,8-hxCDF.
Statistical analysis
Statistical analysis was performed using SAS version 9.3 and JMP version 5.0
software. Basic demographic data were summarized as total numbers and percentage for
categorical variables. Differences of categorical variables were then compared by using
Chi-square tests. Logistic regression was performed using elevated pure tone auditory
thresholds (>20 vs. <=20) at different frequencies as dependent variables, and exposure
22
status (Yucheng vs. referent) as independent variable. Linear regression was performed
using log-transformed pure tone auditory thresholds or log-transformed DPOAEs as
dependent variables, and maternal serum concentrations at pregnancy of
log-transformed maternal PCDFs and marker-PCBs concentrations (at birth) as independent
variable. All regression models were adjusted for potential confounding factors of
hearing loss, such as age, gender, body mass index (Curhan et al., 2013), total
cholesterol (Longnecker et al., 2004), and triglyceride (Chau et al., 2010; Longnecker et
al., 2004).
23